Vector processors are specialized computers that efficiently handle scientific computing tasks using vector instructions. They operate on entire vectors of data simultaneously in pipelines. Key features include common vector operations like addition and multiplication on vectors, vector registers to hold operand and result vectors, and techniques like strip mining and vector chaining to optimize performance on large datasets. Proposed improvements to vector processor design include clustered vector registers and register files to reduce complexity and memory requirements.
3. VECTOR PROCESSOR
• Vector processors are special purpose computers
that match a range of (scientific) computing tasks.
• vector processors provide vector instructions. These
instructions operate in a pipeline .
3
7. OPERATIONS
• Add two vectors to produce a third.
• Subtract two vectors to produce a third
• Multiply two vectors to produce a third
• Divide two vectors to produce a third
• Load a vector from memory
• Store a vector to memory.
7
9. PROPERTIES
• Vector processors reduce the fetch and decode
bandwidth as the number of instructions fetched are
less.
• They also exploit data parallelism in large scientific
and multimedia applications.
• Many performance optimization schemes are used
in vector processors.
• Strip mining is used to generate code so that vector
operation is possible for vector operands whose size
is less than or greater than the size of vector
registers.
9
10. PROPERTIES
• Vector chaining the equivalent of forwarding in
vector processors - is used in case of data
dependency among vector instructions.
• Special scatter and gather instructions are provided
to efficiently operate on sparse matrices.
• Instruction are designed with the property that all
vector arithmetic instructions only allow element N
of one vector register to take part in operations with
element N from other vector registers.
10
11. PROPERTIES
• Based on how the operands are fetched, vector
processors can be divided into two categories - in
memory-memory architecture operands are directly
streamed to the functional units from the memory
and results are written back to memory as the vector
operation proceeds. In vector-register architecture,
operands are read into vector registers from which
they are fed to the functional units and results of
operations are written to vector registers.
11
12. ADVANTAGES
• Data can be represented at its original resolution and
form without generalization.
• Accurate location of data is maintained.
• Efficient encoding of topology, and as a result more
efficient operations.
• Mature, developed compiler technology
• Compact: Describe N operations with 1 short
instruction
12
14. NEW TERMS FOR VECTOR PROCESSORS
• Initiation rate
consuming operands
producing new results.
• Chime
timing measure
vector sequence
ignores the startup overhead for a vector operation.
14
15. • Convoy
is the set of vector instructions
potentially begin execution together in one clock period.
must complete before new instructions can begin.
• vector start-up time
overhead to start execution
related to the pipeline depth
NEW TERMS FOR VECTOR PROCESSORS
15
16. PROPOSED VECTOR PROCESSOR
• CODE (Clustered Organization for Decoupled
Execution) is a proposed vector architecture which
will overcome the some limitations of conventional
vector processors.
16
17. REASONS
• Complexity of central vector register files(VRF) - In
a processor with N vector functional units(VFU),
the register file needs approximately 3N access
ports. VRF area, power consumption and latency
are proportional to O(N*N), O(log N) and O(N)
respectively.
• Difficult to implement precise implementation - In
order to implement in-order commit, a large ROB is
needed with at least one vector register per VFU.
17
18. • In order to support virtual memory, large TLB is
needed so that TLB has enough entries to translate
all virtual addresses generated by a vector
instruction.
• Vector processors need expensive on-chip memory
for low latency.
REASONS
18
19. SOME FEATURES OF CODE
• Vector registers are organized in the form of clusters
in CODE architecture.
• CODE can hide communication latency by forcing
the output interface to look ahead into the
instruction queue and start executing register move
instructions.
• CODE supports precise exception using a history
buffer.
• In order to reduce the size of TLB.
• CODE proposes an ISA level change.
19
20. The Effect of cache design into
vector computers
• Numerical programs
data sets that are too large for the current cache sizes.
Sweep accesses of a large vector
result in complete reloading of the cache
• achieve high memory bandwidth
Register files
highly interleaved memories
• Address sequentiation
20
21. Proposals of cache schemes
• Proposals such as prime-mapped cache schemes
have been proposed and studied. The new cache
organization minimizes cache misses caused by
cache line interferences that have been shown to be
critical in numerical applications.
• The cache lookup time of the new mapping scheme
keeps the same as conventional caches. Generation
of cache addresses for accessing the prime-mapped
cache can be done in parallel with normal address
calculations.
21